Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: MFX after ZINB


From   rgutierrez@stata.com (Roberto G. Gutierrez, StataCorp.)
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: MFX after ZINB
Date   Thu, 13 Feb 2003 16:59:39 -0600

Chris Roebuck <Chris.Roebuck@Advancepcs.com> asks:

> Can someone please explain how one could get POSITIVE marginal effects
> (using MFX) after running a ZINB model where the corresponding coefficient
> estimate is NEGATIVE.

This can occur when the covariate exists in both equations of the model, the
main equation and the "zero inflate" equation.  Without loss of generality,
consider a model with only one covariate (x), which exists in both equations.

The main equation uses the linear predictor: e1 = b0 + b1*x.
The inflation equation uses the linear predictor:  e2 = a0 + a1*x.

In which case, the predicted value used by default in -mfx- is:

                   exp(e1bar)
      nhat =   --------------------
                  1 + exp(e2bar)
               
where e1bar is e1 with x replaced by the mean of the x's and e2bar is similarly
defined.

The marginal effect is the derivative of nhat with respect to x, evaluated at
x = xbar (default for -mfx-).  In this case,

      nhat'(xbar) = exp(e1bar)*exp(e2bar)*(b1 - a1) + exp(e1bar)*b1
                    -----------------------------------------------
                         (1 + exp(e2bar))^2

The denominator is always positive, so that doesn't affect the sign of the
marginal effect.  Note, however, the factor (b1 - a1) in the numerator.  It is
because of this factor that in certain situations you can have a1 and b1
negative, but a positive marginal effect.  This occurs when b1 is sufficiently
greater than a1 so as to "negate" the effect of the second term in the
numerator.

--Bobby
rgutierrez@stata.com
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index